fs/ext4/fsync.c

   1 /*
   2  *  linux/fs/ext4/fsync.c
   3  *
   4  *  Copyright (C) 1993  Stephen Tweedie (sct@redhat.com)
   5  *  from
   6  *  Copyright (C) 1992  Remy Card (card@masi.ibp.fr)
   7  *                      Laboratoire MASI - Institut Blaise Pascal
   8  *                      Universite Pierre et Marie Curie (Paris VI)
   9  *  from
  10  *  linux/fs/minix/truncate.c   Copyright (C) 1991, 1992  Linus Torvalds
  11  *
  12  *  ext4fs fsync primitive
  13  *
  14  *  Big-endian to little-endian byte-swapping/bitmaps by
  15  *        David S. Miller (davem@caip.rutgers.edu), 1995
  16  *
  17  *  Removed unnecessary code duplication for little endian machines
  18  *  and excessive __inline__s.
  19  *        Andi Kleen, 1997
  20  *
  21  * Major simplications and cleanup - we only need to do the metadata, because
  22  * we can depend on generic_block_fdatasync() to sync the data blocks.
  23  */
  24
  25 #include <linux/time.h>
  26 #include <linux/fs.h>
  27 #include <linux/sched.h>
  28 #include <linux/writeback.h>
  29 #include <linux/jbd2.h>
  30 #include <linux/blkdev.h>
  31
  32 #include "ext4.h"
  33 #include "ext4_jbd2.h"
  34
  35 #include <trace/events/ext4.h>
  36
  37 /*
  38  * If we're not journaling and this is a just-created file, we have to
  39  * sync our parent directory (if it was freshly created) since
  40  * otherwise it will only be written by writeback, leaving a huge
  41  * window during which a crash may lose the file.  This may apply for
  42  * the parent directory's parent as well, and so on recursively, if
  43  * they are also freshly created.
  44  */
  45 static int ext4_sync_parent(struct inode *inode)
  46 {
  47         struct dentry *dentry = NULL;
  48         struct inode *next;
  49         int ret = 0;
  50
  51         if (!ext4_test_inode_state(inode, EXT4_STATE_NEWENTRY))
  52                 return 0;
  53         inode = igrab(inode);
  54         while (ext4_test_inode_state(inode, EXT4_STATE_NEWENTRY)) {
  55                 ext4_clear_inode_state(inode, EXT4_STATE_NEWENTRY);
  56                 dentry = d_find_any_alias(inode);
  57                 if (!dentry)
  58                         break;
  59                 next = igrab(dentry->d_parent->d_inode);
  60                 dput(dentry);
  61                 if (!next)
  62                         break;
  63                 iput(inode);
  64                 inode = next;
  65                 ret = sync_mapping_buffers(inode->i_mapping);
  66                 if (ret)
  67                         break;
  68                 ret = sync_inode_metadata(inode, 1);
  69                 if (ret)
  70                         break;
  71         }
  72         iput(inode);
  73         return ret;
  74 }
  75
  76 /**
  77  * __sync_file - generic_file_fsync without the locking and filemap_write
  78  * @inode:      inode to sync
  79  * @datasync:   only sync essential metadata if true
  80  *
  81  * This is just generic_file_fsync without the locking.  This is needed for
  82  * nojournal mode to make sure this inodes data/metadata makes it to disk
  83  * properly.  The i_mutex should be held already.
  84  */
  85 static int __sync_inode(struct inode *inode, int datasync)
  86 {
  87         int err;
  88         int ret;
  89
  90         ret = sync_mapping_buffers(inode->i_mapping);
  91         if (!(inode->i_state & I_DIRTY))
  92                 return ret;
  93         if (datasync && !(inode->i_state & I_DIRTY_DATASYNC))
  94                 return ret;
  95
  96         err = sync_inode_metadata(inode, 1);
  97         if (ret == 0)
  98                 ret = err;
  99         return ret;
 100 }
 101
 102 /*
 103  * akpm: A new design for ext4_sync_file().
 104  *
 105  * This is only called from sys_fsync(), sys_fdatasync() and sys_msync().
 106  * There cannot be a transaction open by this task.
 107  * Another task could have dirtied this inode.  Its data can be in any
 108  * state in the journalling system.
 109  *
 110  * What we do is just kick off a commit and wait on it.  This will snapshot the
 111  * inode to disk.
 112  */
 113
 114 int ext4_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
 115 {
 116         struct inode *inode = file->f_mapping->host;
 117         struct ext4_inode_info *ei = EXT4_I(inode);
 118         journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
 119         int ret, err;
 120         tid_t commit_tid;
 121         bool needs_barrier = false;
 122
 123         J_ASSERT(ext4_journal_current_handle() == NULL);
 124
 125         trace_ext4_sync_file_enter(file, datasync);
 126
 127         ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
 128         if (ret)
 129                 return ret;
 130         mutex_lock(&inode->i_mutex);
 131
 132         if (inode->i_sb->s_flags & MS_RDONLY)
 133                 goto out;
 134
 135         ret = ext4_flush_unwritten_io(inode);
 136         if (ret < 0)
 137                 goto out;
 138
 139         if (!journal) {
 140                 ret = __sync_inode(inode, datasync);
 141                 if (!ret && !hlist_empty(&inode->i_dentry))
 142                         ret = ext4_sync_parent(inode);
 143                 goto out;
 144         }
 145
 146         /*
 147          * data=writeback,ordered:
 148          *  The caller's filemap_fdatawrite()/wait will sync the data.
 149          *  Metadata is in the journal, we wait for proper transaction to
 150          *  commit here.
 151          *
 152          * data=journal:
 153          *  filemap_fdatawrite won't do anything (the buffers are clean).
 154          *  ext4_force_commit will write the file data into the journal and
 155          *  will wait on that.
 156          *  filemap_fdatawait() will encounter a ton of newly-dirtied pages
 157          *  (they were dirtied by commit).  But that's OK - the blocks are
 158          *  safe in-journal, which is all fsync() needs to ensure.
 159          */
 160         if (ext4_should_journal_data(inode)) {
 161                 ret = ext4_force_commit(inode->i_sb);
 162                 goto out;
 163         }
 164
 165         commit_tid = datasync ? ei->i_datasync_tid : ei->i_sync_tid;
 166         if (journal->j_flags & JBD2_BARRIER &&
 167             !jbd2_trans_will_send_data_barrier(journal, commit_tid))
 168                 needs_barrier = true;
 169         jbd2_log_start_commit(journal, commit_tid);
 170         ret = jbd2_log_wait_commit(journal, commit_tid);
 171         if (needs_barrier) {
 172                 err = blkdev_issue_flush(inode->i_sb->s_bdev, GFP_KERNEL, NULL);
 173                 if (!ret)
 174                         ret = err;
 175         }
 176  out:
 177         mutex_unlock(&inode->i_mutex);
 178         trace_ext4_sync_file_exit(inode, ret);
 179         return ret;
 180 }